EdEarl Posted August 23, 2016 Posted August 23, 2016 Phys.org Princeton University researchers have built a new computer chip that promises to boost performance of data centers that lie at the core of online services from email to social media. In recent years companies and academic institutions have produced chips with many dozens of cores; but Wentzlaff said the readily scalable architecture of Piton can enable thousands of cores on a single chip with half a billion cores in the data center. That half billion cores can share a single memory, which a memory fetch to be shared among all those processors; thereby, decreasing total memory fetches and power requirements while improving distributed processing performance. This kind of optimization improves data center cloud performance, and should work well with many common applications and deep learning AI. 1
fred2014 Posted September 9, 2016 Posted September 9, 2016 I rather suspect this one is pie in the sky. The issues surrounding pipelining and parrallel programming make such a device highly innefficient - less so that a single core I would suspect. the notion of "more is better" is a comfy academic fantasy here I think. Not to mention finite access to system resources.
EdEarl Posted September 9, 2016 Author Posted September 9, 2016 I rather suspect this one is pie in the sky. The issues surrounding pipelining and parrallel programming make such a device highly innefficient - less so that a single core I would suspect. the notion of "more is better" is a comfy academic fantasy here I think. Not to mention finite access to system resources. I agree that no one parallel processor can do equally well on all algorithms. However, video processing has yielded to parallel processors with excellent results. Similarly, business data processing with multitudes of users are serviced adequately by the cloud. The parallel requirements for video processing and the cloud are different and the hardware to support them are different. I believe the piton processor is designed for processing algorithms that the cloud currently processes; furthermore, the optimizations they made were designed to improve communications in this environment. I suspect a piton processor would not do as well on video processing as machines made for that task. Whether it is a significant improvement for cloud processing may be tested, and results reported; if so, we shall see. Perhaps I am wrong and it is intended for another kind of parallel environment. Perhaps you are right, their design has no merit.
fred2014 Posted September 12, 2016 Posted September 12, 2016 edEarl - I'd like to be wrong and for this to be a great advance. But I just can't see it. As for "cloud" computing - if I had my way it would be a criminal offence for any business to go anywhere near it.
Recommended Posts